| Commit message (Collapse) | Author | Age |
|
|
|
| |
Show more info when git-cat-file --batch goes wrong
|
| |
|
|
|
|
| |
Call site(s) will come later.
|
|
|
|
|
|
|
|
|
| |
We are going to introduce this option soon. It is not supposed to
actually have very much overall effect, although it will replumb the
way dgit goes about making the source package.
So the tests are just the same as the normal tests only with the
additional option.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
We are going to start handling weirder kinds of git tree, which will
require a kind of `split brain': the dgit-using maintainer's view will
be a git tree which is not a dgit git tree. dgit will convert them
during push.
For this to work we will have to have dgit always (in these cases)
generate the source package itself (with a separate invocation of
dpkg-source). This will involve some dis- and re-entangling of the
way we generate arguments to dpkg-buildpackage.
We can profitably split this up into this pre-patch, which has no
overall functional change for normal users.
We do here provide a new --always-split-source-build option which
allows the new approach to be explicitly requested. This allows us to
use the test suite to not only test that we didn't break any of the
existing building strategies, but also that the new strategies all do
what we expect.
But this option is not really semantically useful for users so we do
not document it.
|
| |
|
|
|
|
|
| |
Otherwise one failure causes the tests to be abandoned, which is not
usually desirable.
|
|
|
|
|
|
| |
Some of the tools we use, it turns out, do not update properly: they
overwrite in place. This can cause inadvertant changes to a tartree
opened for editing.
|
| |
|
|
|
|
| |
Simply HEAD is not displayed by gitk, which is rather a shame.
|
|
|
|
|
| |
When we unmerged some old A+B merged refs, make a note to merge them
again. This saves the user calling gitfetchinfo-merge again.
|
|
|
|
|
|
| |
* Turn on perl warnings
* Use strict
* When removing old merged refs, insist that we aren't handling A+A
|
|
|
|
|
|
|
|
|
|
|
|
| |
The sequence
gitfetchinfo-merge A B
gitfetchinfo A
gitfetchinfo-merge A B
might leave inappropriate A+B refs, because nothing would delete them.
So, have gitfetchinfo A unmerge any A+B or B+A refs. (This does mean
that the user of gitfetchinfo A is now promising that any remotes A+B
or B+A were made with gitfetchinfo-merge.)
|
|
|
|
|
|
| |
We want the commit timestamps to be predictable (ie, to depend on the
source info only), so that we don't show spurious disagreement between
two working trees. So, use the timestamp of the HEAD commit.
|
| |
|
| |
|
|
|
|
| |
Help with comparing different test case git working tree tarballs.
|
| |
|
| |
|
|
|
|
| |
unexpected output
|
|
|
|
| |
--rm-old-changes, or failing that, fail early.
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Instead, insist on a single one. This eliminates a potential for a
variety of surprises, where dgit would upload something different to
the user expected.
We no longer need to behave this way because nowadays the sbuild build
(which was the real use case, because it would generate a .source and
a .ARCH changes as well as the final .multi) renames the intermediate
changes files.
Also --rm-old-changes helps.
|
|
|
|
| |
safety check.
|
|
|
|
|
|
| |
Soon, we are going to be stricter about detecting ambiguity about
which .changes file is intended when different build modes have been
used.
|
| |
|
| |
|
| |
|
| |
|
| |
|
|
|
|
| |
This avoids accidental use of the wrong one (by software, or by users).
|
|
|
|
|
| |
Check that the set of .changes files found is as we expect, before
calling mergechanges.
|
| |
|
|
|
|
|
| |
zealot is very slow and we need to give the other processes time to
rollback and release the lock.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This avoids most date dependencies: unless other measures are taken,
the commits and tags are identical iff they are constructed
identically. This makes the test suite more deterministic.
In drs-push-rejects test, increment them explicitly, so as to make all
of mktag's tags distinct objects.
Otherwise it can generate identical tags (now, always; previously,
only sometimes on fast computers), causing spurious variations in
behaviour (eg, things not being rejected because they do not represent
any change).
|
|
|
|
|
| |
Pass --ch:* and -v options to dpkg-buildpackage when building source.
Fixes bad Perl poetry syntax. Closes:#829121.
|
|
|
|
| |
Closes:#800078.
|
|
|
|
| |
cannot create the destination directory.
|
|
|
|
| |
has been made fatal and aborts the cleanup chain).
|
|
|
|
| |
before running rmtree on it. Closes:#796773.
|
|
|
|
| |
Cosmetic change only.
|
|
|
|
|
| |
... due to sed not finding the log file. Instead, simply tolerate the
absence of the log file.
|
|
|
|
|
|
|
|
|
|
| |
Incompatible change:
* dgit sbuild does not pass -A to sbuild. Consequently the default
build is now simply sbuild's default. With older sbuilds it was
possible to override dgit's -A by passing another option. But this
has been changed recently and now this default setting is very
awkward to change for the dgit user.
|
| |
|
| |
|
| |
|
|
|
|
| |
don't need build-depends.
|
| |
|
|
|
|
| |
can override it. Closes:#796019.
|
|
|
|
|
|
| |
Because dpkg-source -b breaks with single-debian-patch when the .pc is
missing, so this depends on split execution of the source package
build.
|